MiniCluster HTML Report
127.8000
0.3348
18.3000
286.5000
| Tool | minicluster 0.1.0 |
|---|---|
| Timestamp | 2026-02-21T02:58:27.987985 |
| Hardware | CPU |
| Framework | pytorch 2.10.0+cpu |
| Workload | cluster_health_demo_run (training) |
| num_processes | 1.0000 |
|---|---|
| num_steps | 10.0000 |
| batch_size | 16.0000 |
| learning_rate | 0.0100 |
| hidden_size | 128.0000 |
| num_layers | 2.0000 |
| collective_backend | nccl |
| workload | transformer |
| seed | 42.0000 |
| tdp_watts | 150.0000 |
| loss_tolerance | 0.0100 |
| regression_threshold | 5.0000 |
| Throughput (samples/sec) | 127.8000 |
|---|---|
| Final Loss | 0.3348 |
| Total Time (s) | 18.3000 |
| P50 All-Reduce (ms) | 2.6200 |
| P95 All-Reduce (ms) | 3.7100 |
| P99 All-Reduce (ms) | 4.2400 |
| Max All-Reduce (ms) | 4.4300 |
| All-Reduce StdDev (ms) | 0.3900 |
| Power (W) | 286.5000 |
| Performance/Watt | 0.4460 |
| Energy/Step (J) | 2.2400 |
| Temperature (C) | 69.8000 |
| Communication Overhead (%) | N/A |
| Scaling Efficiency (%) | 92.4000 |
4.2400 ms. A heavy right-tail in the histogram is your straggler signal.0.3900 ms. Higher variability can indicate unstable interconnect paths.92.4000%. Below ~90% often means communication or memory bottlenecks.Coverage: all key cluster-health aggregates are present.